Yes, AI companies destroy physical books, and it isn't a rumor: a federal judge confirmed it in a 2025 ruling. Anthropic bought millions of used print books, sliced off the bindings with a hydraulic cutter, scanned every page, and threw out what was left.
Short answer: A federal court confirmed in June 2025 that Anthropic destroyed millions of purchased print books after scanning them for AI training, under an internal program called Project Panama. A judge ruled that specific act legal as fair use. Separately, Anthropic paid $1.5 billion to settle claims over pirated books it also used. If you own books you can't replace, scan them yourself before they end up in someone else's pipeline.
Last updated: August 21, 2026 · By Vishal Swami, Founder & Lead AI Reviewer, AISagely
What actually happened to the books
The documented case is Anthropic, in the copyright lawsuit Bartz v. Anthropic. In my testing of the claims flying around this story, I skipped the aggregator summaries and went to the court filings and the legal writeups tracking them directly, because "AI companies destroy books" is exactly the kind of headline that gets exaggerated in the retelling. It wasn't exaggerated.
Under an internal effort called Project Panama, Anthropic hired the former head of partnerships for Google Books and bought used copies of millions of print titles from secondhand sellers like Better World Books and World of Books. Warehouse workers then ran each book through a hydraulic-powered cutting machine to remove the binding, fed the loose pages through industrial scanners, and discarded or recycled what was left. The goal was a clean, high-volume digital library for training Claude.
In June 2025, Judge William Alsup ruled that this specific act — buying a physical book you're entitled to own, converting it to a digital copy, and disposing of the original — is fair use. He treated it as a "format change," not new copying, since one physical copy simply became one digital copy. Anthropic's separate practice of downloading over 7 million books from pirate sites like Library Genesis for its permanent library was a different matter: the court found no fair-use defense there, since Anthropic never held a legitimate copy to "format-shift" from, according to Loeb & Loeb’s summary of the ruling. That piracy claim is what actually cost Anthropic money: a $1.5 billion settlement, the largest copyright class action in U.S. history, which a federal judge gave final approval on July 20, 2026, per TechCrunch’s coverage of the approval. Roughly 500,000 works are covered, at about $3,000 per work.
So the physical destruction itself was ruled legal. What should worry a book owner isn't that Anthropic broke the law — it's that an AI lab can now legally buy a book, including a scarce or out-of-print one, and permanently remove the physical copy from existence once it's been digitized for a private training set. If you're holding a first edition, a signed copy, a family Bible, or anything a used-book buyer might scoop up in bulk, waiting for someone else to preserve it is a bad bet.
What you'll need
You don't need a warehouse or a hydraulic cutter — the opposite, actually. A smartphone with a decent camera covers most books. Add a free scanning app, a flat, well-lit surface, and something to weigh the pages open without cracking the spine (a couple of paperbacks work fine). For anything genuinely rare, fragile, or over a few hundred pages, a dedicated non-destructive book scanner is worth the cost, because it flattens the curve near the spine optically instead of forcing the book flat. Set aside real time: a 300-page book takes 30–45 minutes to photograph carefully, longer if the paper is brittle.
Step-by-step: digitizing a book without destroying it
1. Photograph or scan each spread
Prop the book open at roughly 100–120 degrees, not flat, to protect old glue and stitching. Use a scanning app's edge-detection rather than your camera roll — it crops and corrects perspective automatically as you go.
2. Flatten and de-skew the pages
This is where the anti-Anthropic approach pays off. A dedicated scanner like the CZUR ET24 Pro uses laser mapping to digitally remove the page curve caused by the binding, so you never cut anything. On a phone, angle the camera slightly and let the app's auto-crop handle the rest; it won't be as clean, but it's free.
3. Run OCR to make the text searchable
A scan is just a picture until OCR turns it into selectable, searchable text. I compared free OCR tools for exactly this in my PixelRead AI OCR review; for scanning something as large as a full book in bulk, an API-based option like Mistral OCR handles hundreds of pages faster than a phone app processing one image at a time.
4. Clean up the OCR output with an AI assistant
OCR mangles old typefaces, footnotes, and yellowed pages more than people expect. Paste a chapter into an AI assistant and ask it to fix obvious character errors without rewriting the author's actual words — see the example prompts below.
5. Build a searchable archive
Once you have clean text, upload it to a tool built for querying documents rather than just storing them. I walk through the setup in how to use NotebookLM, which turns a pile of digitized chapters into something you can actually ask questions of.
6. Back up in two places
One local copy, one cloud copy, minimum. A digitized book that lives on a single phone is one cracked screen away from being gone twice.
Example prompts you can copy
For cleaning raw OCR text without altering the author's wording: > "Here is raw OCR output from a printed book. Fix obvious character-recognition errors (like 'rn' misread as 'm' or '1' misread as 'l'), but do not paraphrase, summarize, or change any actual wording. Flag any sentence you're unsure about instead of guessing: [paste text]"
For catching pages the OCR likely got wrong: > "Scan this OCR text for signs of a bad scan — broken sentences, repeated words, missing punctuation, or nonsense strings. List the line numbers you'd flag for a manual recheck, and explain why for each one: [paste text]"
For generating a quick reference card once a chapter is clean: > "Summarize this chapter in five bullet points and pull out any names, dates, or places mentioned, so I can index it later: [paste text]"
Common mistakes to avoid
In my testing of this workflow, the most expensive mistake is skipping research before you scan. If you're not sure what an old book is actually worth, check it against sold listings and expert sources before you do anything irreversible to the binding — how to use Perplexity for research is a fast way to pull citations on an edition's rarity in a few minutes. Beyond that:
- Don't cut the binding to save time. That's the Anthropic move, and it's the one you're trying to avoid. A slightly worse phone scan beats a permanently disassembled book.
- Don't trust an app's star rating at face value. A 4.8-star scanner app can still have a gutted free tier; I cover how to read those ratings honestly in AI tool ratings: how to read them without getting fooled.
- Don't rely on Microsoft Lens. It's a common recommendation in older guides, but Microsoft retired it: new scans stopped working after March 9, 2026, per Microsoft’s own retirement notice. OneDrive's built-in scanner is the replacement Microsoft points to now.
- Don't let an AI cleanup pass rewrite the text. Tell it explicitly to fix character errors only, not to "improve" the prose — otherwise you've quietly replaced the author's words with a paraphrase.
- Don't skip a second backup. Digitizing a book and keeping the only copy on one device defeats the entire point.
Tools that make this easier
I checked each of these against its own current pricing page rather than a review roundup, since app pricing tends to drift within months of a comparison going live.
| Tool | Best for | Price (checked August 2026) | Destructive? |
|---|---|---|---|
| Adobe Scan | Quick phone scans, free OCR | Free app; Acrobat Pro from $19.99/mo (annual, billed monthly) for advanced editing | No |
| Genius Scan | Fast multi-page batches | Free (Basic); Genius Scan Ultra $39.99/yr for OCR + cloud backup | No |
| CZUR ET24 Pro | Large collections, fragile bindings | $599 (24MP, laser page-flattening, 320 dpi) | No |
| Microsoft Lens | — | Retired; scanning disabled since March 9, 2026 | N/A |
None of these require cutting anything. The CZUR is the only one built specifically for books rather than loose documents, and it's the closest thing to what a library-grade digitization setup does — just without the part where the book disappears afterward.
Frequently Asked Questions
Did an AI company really destroy physical books?
Yes. Anthropic's Project Panama bought millions of used print books, cut off the bindings with a hydraulic cutter, scanned the pages, and discarded the originals. A federal judge confirmed this in the June 2025 Bartz v. Anthropic ruling.
Is it legal for AI companies to destroy books they scan?
For books they legitimately purchased, yes — a court ruled that converting a purchased book to digital and destroying the print copy is fair use. It's a different story for pirated books, which is the part that led to Anthropic's $1.5 billion settlement.
What's the easiest free way to scan a book at home?
A phone scanning app like Adobe Scan or Genius Scan, used spread by spread with the book propped open rather than flattened. Both have functional free tiers for basic scanning and OCR.
Do I need special equipment to digitize a rare or fragile book?
Not strictly, but a dedicated non-destructive scanner like the CZUR ET24 Pro is worth it for anything valuable or brittle, since it optically flattens pages instead of forcing the spine open.
Is Microsoft Lens still a good option?
No. Microsoft retired it; new scans stopped working after March 9, 2026. Use OneDrive's built-in scanner, Adobe Scan, or Genius Scan instead.